I think that it "dilutes" the percentage of the spammy words/content vs. "real" words.
Our filters add various points together for the spammy things (all caps, symbols in words, bad words, etc.) and then compares these points to the non-spammy things. Then it gives a probability that the mail is spam. We set the threshold as to what is blocked.
ie: [HTML_50_60, HTML_FONTCOLOR_BLUE, HTML_FONTCOLOR_GREEN, HTML_FONTCOLOR_UNKNOWN, HTML_FONTCOLOR_UNSAFE, HTML_FONT_BIG, HTML_MESSAGE, IN_REP_TO, LINES_OF_YELLING, LINES_OF_YELLING_2, MSGID_GOOD_EXCHANGE, ORIGINAL_MESSAGE, RCVD_IN_DSBL, TT_UPPERCASE_OK, UPPERCASE_25_50]
spam probability 99.80988%
Subject: Eop V~i_c_o`din - Order Meds From Home Now"
When you have all of the nonsense words at the bottom it makes the spammy stuff seem to be a lower ratio of the total content of the page. Then it sneaks through the spam filters.
And the spam wars go on.....